feat: Add comprehensive performance benchmarking (Phase 4)

Phase 4 of optional enhancements: Performance Benchmarking

**New Files:**
- tests/test_adaptor_benchmarks.py (478 lines)
  - 6 comprehensive benchmark tests with pytest
  - Measures format_skill_md() across 11 adaptors
  - Tests package operations (time + file size)
  - Analyzes scaling behavior (1-50 references)
  - Compares JSON vs ZIP compression ratios (~80-90x)
  - Quantifies metadata processing overhead (<10%)
  - Compares empty vs full skill performance

- scripts/run_benchmarks.sh (executable runner)
  - Beautiful terminal UI with colored output
  - Automated benchmark execution
  - Summary reporting with key insights
  - Package installation check

**Modified Files:**
- pyproject.toml
  - Added "benchmark" pytest marker

**Test Results:**
- All 6 benchmark tests passing
- All 164 adaptor tests still passing
- No regressions detected

**Key Findings:**
• All adaptors complete formatting in < 500ms
• Package operations complete in < 1 second
• Linear scaling confirmed (0.39x factor at 50 refs)
• Metadata overhead negligible (-1.8%)
• ZIP compression ratio: 83-84x
• Empty skill processing: 0.03ms
• Full skill (50 refs): 2.62ms

**Usage:**
./scripts/run_benchmarks.sh

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

pyproject.toml

@@ -206,6 +206,7 @@ markers = [
     "e2e: mark test as end-to-end (resource-intensive, may create files)",
     "venv: mark test as requiring virtual environment setup",
     "bootstrap: mark test as bootstrap feature specific",
+    "benchmark: mark test as performance benchmark",
 ]
 asyncio_mode = "auto"
 asyncio_default_fixture_loop_scope = "function"